What is an evaporite?

"…a rock that was originally precipitated from a saturated surface or near surface brine by processes driven by solar evaporation (Warren, 1999).

Formation of evaporites

Mineral formation during evaporation

Mineral formation during evaporation

Objective of the study

Is it possible to capture these mineralogical variations using orbital multispectral data? Image courtesy (Warren,2011).

Evaporite minerals selected for this study

  1. Anhydrite \(CaSO_4\)
  2. Aragonite \(CaCO_3\)
  3. Calcite \(CaCO_3\)
  4. Dolomite \(CaMg(CO_3)_2\)
  5. Epsomite \(MgSO_4.7H_2O\)
  6. Gypsum \(CaSO_4.2H_2O\)
  7. Halite \(NaCl\)
  8. Magnesite \(MgCO_3\)
  9. Thenardite \(Na_2SO_4\)
  10. Trona \(Na_3H(CO_3)_2.2H_2O\)

Spectra were collected from different sources (RELAB, USGS, JPL, and from personal collection). Total 100 spectral representing 10 Spectra from each category.

Representative spectra from each mineral type

Calculation of absorption band depth

Mineral Spectra

Anhydrite

Aragonite

Calcite

Dolomite

Epsomite

Gypsum

Halite

Magnesite

Thenardite

Trona

Mean band depths of all mineral spectra

Most interested wavelength regions based on the 100 evaporite Spectra

Selected multispectral sensors

  1. Landsat 5 (NASA)
  2. Landsat 7 ETM+ (NASA)
  3. Landsat 8 OLI (NASA)
  4. EO-1 ALI (NASA)
  5. Sentinel-2 (ESA)
  6. ASTER (NASA + JAXA)

Spectral response function of multispectral sensors

Landsat 5

Landsat 7

Landsat 8

Sentinel-2

ALI (Advanced Land Imager)

ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer)

Finding the best band index

7 band indices were calculated.

  1. AngleA = Angle between the overall spectral trend and horizontal plane.
  2. AngleB = Angle created by three nearby bands centered at ~ 550 nm wavelength (660 nm for ASTER)
  3. AngleC = Angle created by three nearby bands centered at ~ 850 nm wavelength
  4. AngleD = Angle created by three nearby bands centered at ~ 1650 nm wavelength

Additional band indices for ASTER data,

  1. AngleE = Angle created by three nearby bands centered at ~ 1650 nm wavelength
  2. AngleF = Angle created by three nearby bands centered at ~ 2205 nm wavelength
  3. AngleG = Angle created by three nearby bands centered at ~ 2330 nm wavelength

Learning Vector Quantization (LVQ) model was used to estimate the feature importance. k-fold cross validation method was adopted (k = 10, 3 repeats)

3 Machine learning methods were used,

  1. RandomForest (rf)
  2. Neural Network (nnet)
  3. Support Vector Machine (svmRadial)

k-fold cross validation method was adopted (k = 10, 3 repeats)

Importance of selected band indices: Landsat 5

Performance of selected Machine Learning models: Landsat 5

Importance of selected band indices: Landsat 7

Performance of selected Machine Learning models: Landsat 7

Importance of selected band indices: Landsat 8

Performance of selected Machine Learning models: Landsat 8

Importance of selected band indices: Sentinel-2

Performance of selected Machine Learning models: Sentinel-2

Importance of selected band indices: ALI

Performance of selected Machine Learning models: ALI

Importance of selected band indices: ASTER set 1

Performance of selected Machine Learning models: ASTER set 1

Importance of selected band indices: ASTER set 2

Performance of selected Machine Learning models: ASTER set 2

Concluding remarks

  1. ASTER sensor was the best multispectral sensor to identify the given evaporite minerals.
  2. This also shows the importance of SWIR bands for mapping minerals.
  3. AngleF (~ 2200nm) and AngleG (~ 2300nm) were the best Spectral indices to map given evaporites using aster data.

Notes

The entire study was done using Rstudio.

This study only considered the spectral resolution. The spatial resolution, radiometric resolution, and Signal/Noise ratio of the different sensors were not considered.

Acknowledgements

Many thanks to the R Core Team, and Rstudio for making them as free and open source, all the R library developers for their effort and contribution, and for RELAB, USGS, and JPL for making spectra accessible.

Thank you…….

Thank you for your attention.. www.gayantha.net